一句话:提出 Plan4Code,结合软件工程本体与思维链提示,让 LLM 驱动的代码变更更系统、可解释。
该博士研究探讨如何将软件工程本体与思维链(Chain-of-Thought, CoT)提示相结合,使基于大语言模型(LLM)的代码变更过程更加系统化且具备可解释性。作者提出 Plan4Code 方法,利用 LLM 对代码变更进行规划、执行与解释。研究内容包含三部分:(i) 面向遗留软件的 AI 技术文献综述;(ii) Plan4Code 架构的设计与实现;(iii) 实证评估,以检验其在自动化并改进代码变更方面的有效性。整体目标是通过本体提供领域知识约束、通过思维链提供推理轨迹,从而提升 LLM 代码变更的可控性与透明度。
This doctoral research investigates how software engineering ontologies combined with Chain-of-Thought (CoT) prompting can make LLM-based code changes more systematic and explainable. I propose Plan4code, a novel approach that leverages Large Language Models (LLMs) to plan, execute, and explain code changes. The research comprises (i) a literature review on Artificial Intelligence (AI) techniques for legacy software; (ii) the design and implementation of the Plan4Code architecture; and (iii) an empirical evaluation to assess its effectiveness in automating and improving code changes.